Vision-Based Jigsaw Puzzle Solving with Robotic Arm

碩士 === 元智大學 === 電機工程學系甲組 === 107 === For solving the jigsaw puzzle, when the original image can be accessed. First the SIFT algorithm is used to extract and match the feature points between each piece of the puzzle and the original image. Then, the RANSAC (RANdom Sample Consensus) is used to elimina...

Full description

Bibliographic Details
Main Authors: Chien-Liang Lu, 呂建良
Other Authors: Huang-Chia Shih
Format: Others
Language:zh-TW
Published: 2019
Online Access:http://ndltd.ncl.edu.tw/handle/47we8d
id ndltd-TW-107YZU05442014
record_format oai_dc
spelling ndltd-TW-107YZU054420142019-11-07T03:39:34Z http://ndltd.ncl.edu.tw/handle/47we8d Vision-Based Jigsaw Puzzle Solving with Robotic Arm 基於電腦視覺之拼圖重組並以機械手臂實踐之 Chien-Liang Lu 呂建良 碩士 元智大學 電機工程學系甲組 107 For solving the jigsaw puzzle, when the original image can be accessed. First the SIFT algorithm is used to extract and match the feature points between each piece of the puzzle and the original image. Then, the RANSAC (RANdom Sample Consensus) is used to eliminate incorrect matching pairs. By using the result of RANSAC, the transformation matrix(TM) between the piece of puzzle and the original image can be obtained. The TM contains the coordinates and the rotation angle information. Finally, the TM parameters are applied to the robotic arm to solve the puzzle. However, due to the working space of the robotic arm is limited, we design a pre-rotation algorithm of the robotic arm to reduces running time. Compared with the method of the histogram based full search, the feature point matching method can greatly reduce almost 88.5% computation time. Otherwise, without the original image, the four edges color information in all piece of puzzle is extracted. To calculate the similarity of adjacent pieces by the similarity formula, and apply the concept of Hausdorff distance, the best piece of the puzzle can be obtained to form the initial combination. Through the main track, second combination, and third combination to complete the puzzle reconstruction. Based on the experimental results, among the 25 test images, the average success rate reaches 87.1% when each image cut into 35 pieces, and 17 of them are 100% accuracy. When cut into 70 pieces, 77.8% is obtained in average, and 13 of them are 100% accuracy. Referring to the experimental results, In the case of with the original image, the proposed method eliminates the requirement to adjust the SIFT threshold and it can reduce the calculation time to complete the puzzle. Without the original image, it can complete the puzzle without presetting the position of any piece of puzzle. The proposed method can be applied to many other fields, such as industrial automation, robot vision, banknote rebuilding or archaeological restoration. Huang-Chia Shih 施皇嘉 2019 學位論文 ; thesis 45 zh-TW
collection NDLTD
language zh-TW
format Others
sources NDLTD
description 碩士 === 元智大學 === 電機工程學系甲組 === 107 === For solving the jigsaw puzzle, when the original image can be accessed. First the SIFT algorithm is used to extract and match the feature points between each piece of the puzzle and the original image. Then, the RANSAC (RANdom Sample Consensus) is used to eliminate incorrect matching pairs. By using the result of RANSAC, the transformation matrix(TM) between the piece of puzzle and the original image can be obtained. The TM contains the coordinates and the rotation angle information. Finally, the TM parameters are applied to the robotic arm to solve the puzzle. However, due to the working space of the robotic arm is limited, we design a pre-rotation algorithm of the robotic arm to reduces running time. Compared with the method of the histogram based full search, the feature point matching method can greatly reduce almost 88.5% computation time. Otherwise, without the original image, the four edges color information in all piece of puzzle is extracted. To calculate the similarity of adjacent pieces by the similarity formula, and apply the concept of Hausdorff distance, the best piece of the puzzle can be obtained to form the initial combination. Through the main track, second combination, and third combination to complete the puzzle reconstruction. Based on the experimental results, among the 25 test images, the average success rate reaches 87.1% when each image cut into 35 pieces, and 17 of them are 100% accuracy. When cut into 70 pieces, 77.8% is obtained in average, and 13 of them are 100% accuracy. Referring to the experimental results, In the case of with the original image, the proposed method eliminates the requirement to adjust the SIFT threshold and it can reduce the calculation time to complete the puzzle. Without the original image, it can complete the puzzle without presetting the position of any piece of puzzle. The proposed method can be applied to many other fields, such as industrial automation, robot vision, banknote rebuilding or archaeological restoration.
author2 Huang-Chia Shih
author_facet Huang-Chia Shih
Chien-Liang Lu
呂建良
author Chien-Liang Lu
呂建良
spellingShingle Chien-Liang Lu
呂建良
Vision-Based Jigsaw Puzzle Solving with Robotic Arm
author_sort Chien-Liang Lu
title Vision-Based Jigsaw Puzzle Solving with Robotic Arm
title_short Vision-Based Jigsaw Puzzle Solving with Robotic Arm
title_full Vision-Based Jigsaw Puzzle Solving with Robotic Arm
title_fullStr Vision-Based Jigsaw Puzzle Solving with Robotic Arm
title_full_unstemmed Vision-Based Jigsaw Puzzle Solving with Robotic Arm
title_sort vision-based jigsaw puzzle solving with robotic arm
publishDate 2019
url http://ndltd.ncl.edu.tw/handle/47we8d
work_keys_str_mv AT chienlianglu visionbasedjigsawpuzzlesolvingwithroboticarm
AT lǚjiànliáng visionbasedjigsawpuzzlesolvingwithroboticarm
AT chienlianglu jīyúdiànnǎoshìjuézhīpīntúzhòngzǔbìngyǐjīxièshǒubìshíjiànzhī
AT lǚjiànliáng jīyúdiànnǎoshìjuézhīpīntúzhòngzǔbìngyǐjīxièshǒubìshíjiànzhī
_version_ 1719287915541954560